Sleep, melatonin & circadian rhythm

The master clock is the suprachiasmatic nucleus (SCN), a pair of small nuclei in the hypothalamus sitting directly above the optic chiasm. It keeps time, and it distributes that time to the rest of the brain and body ^[1]. Left in constant conditions it keeps running, which is what makes it a genuine clock rather than a response to daylight.

Its period is not exactly 24 hours, and the figure was misreported for decades. Older estimates put the human period anywhere from 13 to 65 hours, but those studies let participants control their own lighting, which unknowingly resets the clock being measured. Under controlled dim light, the intrinsic period averages 24.18 hours, tightly distributed, with no meaningful difference between young and older adults ^[3]. Being slightly over 24 hours means the clock must be nudged earlier every day, which is what morning light does.

The light input is more specific than it sounds. Rods and cones are not required for it. A small population of retinal ganglion cells contains its own photopigment, melanopsin, and depolarises in response to light even with all rod and cone input pharmacologically blocked; the sensitivity, spectral tuning and slow kinetics of that response match the properties of circadian entrainment ^[4]. These cells are most sensitive to short-wavelength blue light, they integrate over minutes rather than milliseconds, and they project directly to the SCN. This is why light is the dominant sleep lever, and why blind people with intact ganglion cells can still entrain.

Inside every clock cell is a transcription-translation feedback loop: a set of clock genes whose protein products shut off their own transcription, taking about a day to complete a cycle ^[5]. Because those loops exist in liver, muscle, fat and immune cells too, the SCN's job is largely coordination, and the practical consequence is internal misalignment: eating, light and activity can pull peripheral clocks in a different direction from the central one, which is the mechanism behind shift work and jet lag doing damage beyond simple sleep loss.

Sleep is governed by two processes that are worth separating because interventions act on one or the other and almost never both.

Process S is homeostatic sleep pressure. It builds monotonically the longer you are awake and dissipates during sleep. Its best-characterised chemical correlate is extracellular adenosine, a byproduct of the brain's energy use that accumulates across the waking day and promotes sleep by acting at adenosine receptors on wake-promoting neurons. Caffeine works by blocking those receptors, which is why it opposes sleepiness without addressing the underlying pressure; see adenosine receptors. The pressure is still there, and it is still accumulating.

Process C is the circadian signal from the SCN. It is not a simple sleep drive: across the ordinary day it acts as a wake-promoting signal that rises through the afternoon and evening to oppose the accumulating pressure, then withdraws at night. The switch itself is a mutually inhibitory flip-flop between wake-promoting monoaminergic and cholinergic nuclei and sleep-promoting neurons in the preoptic area, stabilised by orexin neurons whose loss causes narcolepsy ^[1]. Flip-flop circuits change state fast and resist intermediate states, which is why sleep onset is comparatively abrupt.

Almost every common sleep complaint falls out of this. Lying awake at midnight after a 3pm nap is low Process S. Being wide awake at 2am after a flight is Process C in the wrong place. Waking at 5am unable to return to sleep is pressure exhausted before the clock says morning. The two need different fixes: pressure responds to time awake and to activity, timing responds to light and to melatonin, and a sedative addresses neither.

Sleep is not uniform. It cycles roughly every 90 minutes between NREM stages, including deep slow-wave sleep, and REM, with slow-wave sleep concentrated in the first half of the night and REM in the second. That distribution matters practically: cutting sleep short at either end removes different things.

Memory consolidation is the best-documented job. The modern account is active rather than passive: recently encoded representations are reactivated during slow-wave sleep and redistributed from temporary hippocampal storage into long-term cortical networks, with subsequent REM apparently stabilising the transformed memory. Slow-wave sleep in particular has moved from a supporting role to the centre of this account, and the waking brain is now understood as optimised for encoding while the sleeping brain is optimised for consolidation ^[6].

Clearance is the newer and more contested finding. In live mice, natural sleep and anaesthesia were both associated with a roughly 60% increase in the volume of interstitial space between cells, sharply raising convective exchange between cerebrospinal and interstitial fluid and roughly doubling the clearance of injected beta-amyloid ^[7]. That is the source of the popular claim that sleep washes the brain. The caveat belongs with it: magnitude and mechanism have both been challenged by later work using different methods, and this is an active argument rather than settled fact.

Sleep also anchors endocrine rhythms, which is why sleep debt shows up in blood. Six nights at four hours in bed raised evening cortisol, lowered glucose tolerance and thyrotropin, and increased sympathetic activity in healthy young men ^[14]; see the HPA axis and cortisol.

The practical upshot for anything sedating: unconsciousness is not sleep. Alcohol and older sedative hypnotics reliably shorten sleep onset while suppressing slow-wave sleep, REM, or both, which is how a full night can leave a person unrefreshed.

Melatonin is secreted by the pineal gland under SCN control, rising in the evening darkness and suppressed acutely by light. Its job is to tell the body's tissues what time the clock thinks it is. Exogenous melatonin can act as a mild soporific, as a chronohypnotic that opposes the circadian wake drive, and as a chronobiotic that shifts the phase of the clock itself, and the third of those is where it is genuinely strong ^[2].

Shifting the clock is done by timing, not amount. A phase response curve maps how much and in which direction a dose moves the clock depending on when it is taken. For 0.5 mg, the largest phase advances came from taking it in the afternoon, roughly 2 to 4 hours before the dim light melatonin onset, which is about 9 to 11 hours before the midpoint of sleep; delays peaked soon after wake time. The optimal administration time is later for the lower dose than for 3.0 mg, and critically, when each dose is given at its own optimal time, both produce advances and delays of similar size ^[8]. More is not more. A large dose at bedtime is close to the least useful point on the curve for shifting a delayed schedule.

As a sleep aid in ordinary insomnia, the honest number is small. A meta-analysis of 19 trials in 1,683 participants found melatonin reduced sleep onset latency by about 7 minutes, increased total sleep time by about 8 minutes, and modestly improved subjective sleep quality. All were statistically significant; all are much smaller than prescription hypnotics; and the effect did not appear to fade with continued use ^[9]. The reviewers' framing is worth keeping: the absolute benefit is smaller than pharmacological alternatives, and the case for it rests on a benign side effect profile.

There is a supply problem worth knowing about. Analysis of 31 commercial melatonin supplements found actual content ranging from 83% below to 478% above the label, with lot-to-lot variation within one product as high as 465%, and no relationship to manufacturer or product type. Eight of the products also contained serotonin, at 1 to 75 micrograms ^[10]. A person carefully titrating a 0.3 mg dose from a product with that error range is not doing what they think they are doing.

The straightforward uses are jet lag, delayed sleep phase and shift work, in small doses at a computed time. The misuse is a large bedtime dose taken indefinitely for general insomnia, where the expected effect is minutes and the timing is wrong for shifting phase.

The foundation is behavioural and it genuinely outperforms the pharmacology. Consistent wake time anchors the clock better than a consistent bedtime does, because wake time controls the light exposure that entrains it. Morning outdoor light is orders of magnitude brighter than indoor lighting. A dark, cool room helps, since core temperature has to fall for sleep onset. Cognitive behavioural therapy for insomnia outperforms hypnotics on durable outcomes and is the first-line recommendation in every major guideline.

Two specific behavioural findings are worth quoting because people underestimate both. Caffeine at 400 mg disrupted sleep significantly relative to placebo even when taken six hours before bed, in a study measuring sleep objectively as well as by report; the authors' conclusion was that the six hour buffer is the empirical minimum, not a comfortable margin ^[12]. And reading on a light-emitting screen in the hours before bed, compared to a printed book, lengthened sleep onset, reduced evening sleepiness, suppressed melatonin, delayed circadian phase and reduced next-morning alertness ^[11]. Both effects are larger than anything most supplements achieve.

On the compound side the evidence is thinner than the shelf space suggests, and the honest summary is in the table. Melatonin for timing. Glycine at 3 g before bed reduced next-day fatigue and improved psychomotor vigilance in sleep-restricted volunteers, in a small industry study ^[13]. L-Theanine is calming without sedation and its sleep data is mostly indirect. Magnesium matters if intake is low and is unremarkable if it is not. Apigenin, the chamomile flavonoid, has plausible benzodiazepine-site activity and almost no controlled human sleep data.

The stronger drugs trade sleep quality for quantity. Diphenhydramine builds tolerance to its sedative effect within days and carries anticholinergic load. GABAergic hypnotics such as zolpidem work but carry the tolerance and dependence profile of that whole class. Ethanol shortens onset then fragments the second half of the night. Orexin antagonists such as suvorexant are a newer mechanism, blocking the wake-stabilising signal rather than adding sedation.

What is not known deserves saying. There is no established optimal amount of slow-wave sleep to aim for, and no consumer wearable is validated against polysomnography for staging it, so most "deep sleep" numbers are estimates from movement and heart rate. Whether improving sleep in a healthy person changes long-term disease risk has been shown by association, not by randomisation. And the glymphatic mechanism, the most-quoted reason to sleep, is real in its original measurements and still argued about in magnitude ^[7]. Not medical advice.